From the Burrow

TaleSpire Dev Log 130

2019-11-19 23:06:10 +0000

Today I carried on working on the scripting implementation.

Each script has a small section of memory that it will be able to store values to, my first job of the day was finishing of that data store. A few things were a little difficult to debug as VS2017’s debugger doesn’t show what data is on the other end of a pointer.

After finishing I found out that VS2019 does have this feature so I spent a couple of hours fighting to get that upgraded and playing with Unity. If you try this yourself remember:

  • Don’t let VS install unity again, the version is old. Search for ‘Unity’ in the components section of the installer and find the tools “unity tools for visual studio” component
  • Open Unity’s package manager and make sure “Visual Studio Code Editor” is up to date
  • If VS2019 isn’t appearing in Preferences -> External Tools -> External Script Editor then click browse and find something similar to C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\Common7\IDE\devenv.exe
  • If like me you only use VS for debugging and have a different editor for programming you will need to update omnisharp (or your equivalent) that and probably your editor integration too.
  • Restart everything :p Just for safety

After all this, I hooked up the scripting system so that the board asset arrays now receive the references into Spaghet’s data store. All is working according to plan there.

The next thing I want to do is look at moving Spaghet into its own package so I can share that between TaleWeaver (the asset prep and modding tool) and TaleSpire itself.

That’s gonna take a little coordination with the rest of the team so during that I’m going to have a look at making the Unity component that lets you connect resources to your scripts. Here is the old (very unfinished) one that held the LUA scripts.

script assets

The important part, for now, is the bit marked in red. You need to be able to add resource slots so you can drag in the animator or sound effect you are interested in. Then those resources can be referred to from the script.

Another potentially nicer way to handle it would be have the slots on the state machine nodes themselves.. hmm actually there might be a nice way to do that as the nodes are made with Unity’s imgui elements. I’ll look into that tomorrow.

Whatever approach I use for beta, I’ll then need to look for this element when the asset is first loaded in TaleSpire and submit the scripts to Spaghet.

Slowly slowly all these threads are weaving together.

More tomorrow.


TaleSpire Dev Log 129

2019-11-19 01:07:45 +0000

Today I’ve continued working on scripting. My main goal has been to work out the most efficient way to handle the allocation and setup of the private data that each instance of a script is allowed to store data into.

There can be large numbers of tiles made in a single action and so making sure I’m no performing unnecessary copying of data has been important. On the flip side many creation events are of only a single tile so whatever we have needs to make sense at that scale too.

As I progressed with this I kept feeling that the native collections I had available, whilst great, were not ideal for this task so I decided to look into how to write my own. I really wanted something chunked like NativeChunkedArray but with

  • faster indexing
  • no deallocation of backing chunks unless explicitly requested
  • An api more focused on being backing store for the data rather than being focused on working on an element by element basis

The official documentation is both limited and also kind of out of date so I’d recommend starting with this fantastic article series by Jackson Dunstan. In fact if you are considering doing any work with Unity’s new DOTS systems, trawl that site, it’s a goldmine.

Pulling apart his examples and getting to grips with Unity’s safety system took the rest of the evening, but it was definitely worth it. I’ve now got the base of a much more focused chunked store that I can ensure provides exactly what the scripting system needs.

That’s all for tonight.


TaleSpire Dev Log 128

2019-11-16 16:56:16 +0000

Update time! Progress has been very good.

@Ree has kept on hammering away at the building tools from the last update. He’s also been handling lots of behind-the-scenes organizational stuff which whilst not exciting to blog about is as critical to TaleSpire happening as anything else we do.

Jason has been chatting to loads of the backers who pledged at the ‘Help design a ___’ levels and sculpting is in full swing. Very exciting!

I’ve finally broken through the wall of fiddly details that was plaguing my board sync implementation and it’s looking pretty good now. This has freed me up to look at some other tasks that are on my todo list so I’m gonna ramble about that for a bit in this post.

First off I went back to Fog of War. Basically, the task is this:

  1. Take a 16x16x16 grid
  2. In every cube write ‘true’ if there is meant to be fog there and ‘false’ if not
  3. Now take this info and make a 3D mesh that contains all the cells marked ‘true’

And I wanted to make sure we had step 3 worked out.

Now there are a couple of points to keep in mind:

  • The mesh we need to make needs to conform fairly closely to the grid as we don’t want to see glimpses of anything we shouldn’t
  • We don’t need to make this look like fog, we just need a mesh we can start working with, we can do all kinds of polygonal and shader effects on top of a simple mesh that will look much more fog-like.

As a great example check out this tweet from @Ed_dV !

Those clouds are spheres, with lots of magic on top (if you watch the whole thing it shows the spheres without the magic).

Given those two points, I decided a good source of wisdom would be the kinds of algorithms that Minecraft-like games often use. Luckily for me Mikola Lysenko of has written some amazing articles on these techniques along with WebGL demos and source code, what a star! This gave me a huge boost and along with a few implementation tips from other sources I was able to put together a ‘Greedy Mesher’ (the name of one of the techniques) in Unity. Now, this next picture is not meant to be clouds, it was just a test to make sure it was working.

greedy mesher

For the coders out there, I also made this implementation using Unity’s job system which allows me to run the mesh generation jobs in parallel across all cores. It’s pretty speedy considering how little effort was required.

That is all on the subject of fog for now but we will be back with more in future.

Next on my list of concerns was interactable Tiles and modding.

We have a bunch of interactable tiles in TaleSpire and this is only going to increase as we open up the ability for the community to make them.


A real pain point when you make a user-driven content game like TaleSpire is that you have no way to know when or where someone will just suddenly throw tonnes of extra stuff for your game to do.

For example, take that side-table in the gif above. There is nothing to stop you dragging out a whole load of tables and the numbers get large FAST. Remember that if you drag out a square with 32 tiles down one side then that is over 2000 tiles that your game suddenly has to handle. They all need to appear on all the other player’s machines quickly and they all need to be interactable immediately.

That means all 2000 of them are running their own little scripts. How do you ensure that they don’t start crippling your game?

Now we naturally knew this was coming and previously we had been making small fast components that would then be wired up using a LUA script (that’s another programming language for those not into this stuff :] ).

It worked but I was still concerned about how it would scale.

There was another thing. Like I said we were using LUA in a weird way, as a way of setting stuff up. I started thinking that that might be the worst of all worlds for users as, for those who like LUA, there are lots of things they would be told not to do and for those who hate LUA, well they hate it :p

Were there other problems? You’re damn right there were!

As mentioned before, we need to keep all of these tiles in sync on everyone’s machines and so the idea of scripts just being able to fire off messages whenever they liked was a nightmare, it could really make the game unstable if done wrong.

Now, this sounds tautological but there are two kinds of state for a tile: State that matters to the story and state that doesn’t. For example, whether a fire is lit or not matters, the exact positions of every smoke particle does not. As long as it roughly conforms all is well.

So it would be good if we can separate these kinds of scripts out from each other so we have a shot at being able to enforce sensible behavior over the network.

So after pages and pages of notes and a lot of coffee, I decided it would almost be worth making our own compiler that suited our needs but we needed

Now, this bit gets technical so feel free to skip to the row of asterisks!


I wanted something that

  • has a simple execution model.
  • was focused on only the tasks we need.
  • run on multiple cores
  • used no GC
  • was reasonably fast

Here’s the approach I went for:

0. I decided to start with only 4 data types: int, float, int3, and float3

1. I use a node graph as the code representation. I’m currently using the excellent XNode but if/when Unity’s new graph UI stabilizes I will probably move to that.

I added code to detect cycles and compute the correct types at each node

2. Walk the graph and generate a tree of IR nodes. This is mainly for convenience but in the conversion, we also pick correctly typed IR nodes in a few specific places.

3. Walk the IR nodes to compute the layout for the stack

4. Walk it again and generate a bytecode that represents the program

5. Do a little cleanup pass and kick the result out as a NativeArray<byte>

6. Write a Job that takes the instruction array, an array of private state and a few other bits and run our little bytecode in parallel.

7. Marvel at the power of coffee

It took 18 hours to get the first version up and running and the results were promising.

Here is a picture of some nonsense ‘code’


The graph above (whilst being gibberish) can be run for 10000 tiles in around 2ms, which is way slower than it will be but it’s acceptable for a first pass. It told me the approach was worth more work so I took Friday to flesh it out a bit.

Oh, and the language is called Spaghet!

With that we have the kind of script that can be used for the ‘not story critical stuff’ but we still need the other scripts too.

To that end, I’m now working on how to script the state machines that will run the important stuff. Here is a little picture of the prototype I’m currently working on:


Naturally, the look of the graph and the nodes available will be improved too.

Soooo yup, it’s been a good, busy week. I’m going to be working on this and all the code that holds it together all this week. I’m hoping to have the major plumbing done by Wednesday though.

Cutting Tiles Take 2

2019-10-06 19:44:47 +0000

This weekend Ree and I also took another look at the tile cutaway effect. This one is super important not just to reveal the part of the building you are looking into, but also to cut down walls that would get in the way of controlling your creatures.

Please note: All the following screenshots are not from TaleSpire or even Unity, they are from some test code I was using to play with cutaway effects. It’ll look much nicer in TaleSpire :p

We had prototyped this effect before. Here is a sphere we have chopped the top of using the technique.


The basic approach was that we discard all the fragments that are above the cutaway and then texture the backfaces as if they were positioned at the flat cutaway plane.

However, we saw an issue when the meshes of the models were intersecting (we also saw this in some of the assets on the Unity store for drawing cutaways)


So what can we do? Well if we look at this with the power of MS Paint this is what we have.


So let’s draw some lines from an imaginary camera point to a few points across the cutaway spheres. Then let’s make a score for each line if the line crosses a backface you add 1 point, for a frontface you subtract 1.


Neat, notice how if the score is above zero then it means you are seeing through the cutaway. Also note that because it’s addition and subtraction, we can do it in any order.

We can perform this check for every pixel really easily. Let’s render all the backfaces first and additively blend all the fragments. Then let’s render all the front faces into the same buffer subtractively blending.

Then we have an image which is the mask for the cutaway area.


At this point, we have effectively solved the original problem. We can now render the tiles (throwing away the stuff above the cutaway) and then render the cutaway using the mask we have made.

Note: you could definitely do some nice stuff with stencil buffers to speed all this up.

And there it is, a simple cutaway handling intersections.


Now this may be the version we use in TaleSpire or it may not, we still need to work on it some more and see how it performs once it’s in. But either way, it’s neat :)

Alrighty, that’s all for now.


TaleSpire Dev Log 127

2019-09-17 08:24:47 +0000

Recently I’ve been spending most of my time working on undo/redo support and general data model work inside TaleSpire. Because of that, I thought I’d take a post to ramble a little about undo/redo and how a ‘simple’ feature can grow when you have a few constraints.

This is not the exact way we are doing things in TaleSpire, but it’s an exploration of an idea I went through whilst looking at the feature.

In this post, I’m only going to talk about placing/deleting tiles rather than creatures to keep things simpler.

Chapter 1 - Building basics

So lets first have a look at a little building:

Players can do a few things:

  • They can place a single tile
  • They can drag to place multiple tiles
  • They can delete a single tile
  • They can select and then delete multiple tiles
  • They can copy and paste slabs of tiles

We are going to refer to placing tiles as an add operation and deleting tiles as a delete operation

There is no moving of tiles in TaleSpire, anything that looks like a move is actually a delete operation followed by an add operation. So cut and paste work like this. However, for now, we will ignore copy/paste and just look at the fundamental operations.

What can we say we know about add operations:

  1. The building tools should ensure that the new tile/s do not intersect any existing geometry
  2. The building tools should ensure that no two tiles in a drag add operation intersect.

For now, we will just leave these points here but we will refer back to them later. One thing of note though is that the ‘does not intersect’ property is something we want to try and maintain. Due to undo/redo it’s pretty easy for people to violate this[0] but the rest of the system should try to maintain this as much as possible.

All right, so our building tools let the player add and delete tiles, but we have to maintain this state in memory. Our world (the board in TaleSpire) can be far too big to fit in memory all at once so we divide the world up into zones which we can load in and out on demand. Because of this, we will want zones to be as independent as possible and so each zone is going to maintain it’s own tile data.

An add or delete of a single tile seems like a simple place to start, it should only affect one zone right? Well, it depends on the size of the tile. Let’s say our zones are 16x16x16 units in size and that the minimum size of a tile is 1x0.25x1 (where the 0.25 is the vertical component), well then if a single tile is 40x1x40 then it could be overlapping nine different zones. Even a humble 2x2x2 tile can overlap eight zones if it is in the corner. For the sake of sanity let us say that a tile can be no bigger than half the size of the zone (8x8x8), this way it can at most overlap eight zones, just like the 2x2x2 tile. We will also leave this issue here for now as we will have enough to think about, however, keep in mind how many little things we are ignoring for the sake of the readability of the article, all of these will need to be resolved before your game ships :p

Back to adding tiles! When tiles are placed we need to store data about them, things like the kind of tile it is, the position, rotation, etc. We will just talk about a tile’s data as if it’s a single thing for simplicity[1].

First player0 drags out 3 tiles.

tile-data = [add(* * *)]

tile-data is some linear collection holding the tile data and add is an object holding three tiles represented by asterisks (*) in this case.

Ok so they add 2 more tiles

tile-data = [add(* * *), add(* *)]

Neat. Now let’s look at undo. Well, that’s pretty simple right now as we have been storing our tiles in the order they happened, let’s just drop that most recent add.

tile-data = [add(* * *)]

Hmm, but now we need to support redo, it’s a shame we just threw away the data on the undo. What if we didn’t, what if we had a little store of things that have been ‘undone’. That means after the undo the data would look like this

tile-data = [add(* * *)]
undone-data = [add(* *)]

and now redo is simple too. We pop the data off of the undone-data stack and push it back onto tile-data.

tile-data = [add(* * *), add(* *)]
undone-data = []

So far so groovy. But now let’s think about deleting tiles. Well, that doesn’t map to our data layout as neatly. Our tile data is organized in chunks that are ordered by when they happened in history, not by where they are in the world. Our zones do offer a limited amount of subdivision of the world so currently, this means we need to visit each zone the selection intersects and then scanning through all the tiles in those zones to see which are in the selection. [2]

So about those zones.. how many are going to be involved? Let’s look at this 2d version.

2d selection example

Notice how, as the size increases the number of zones we need to consider is going up by the width x height and it’s more painful in 3D as it’s increasing by width x height x depth, so roughly cubing with respect to edge length. This is going to be quite a lot of zones to check. But how many tiles can a zone hold? Well a zone is 16x16x16 but a tile can be as small as 1x0.25x1 then that is 16x64x16 tiles 16384tiles…ugh this is getting to be a lot of work.

But a ray of hope appears, notice the hatched areas in the diagram above. Those are zones that are completely enclosed by the selection. Naturally, this means a delete operation will delete all the tiles within so we have an opportunity for a big optimization. For any zone that is completely enclosed we just remove all tiles, and we only compare the selection with the tiles themselves for zones that are only partially intersected by the selection. In general, this means the outer ‘surface’ of zones. The number of zones on the surface of the selection grows roughly in the order of (x^3 - max(x-2, 0)^3) which is much more manageable.

The situation has improved then but we still need to apply the operation to all those zones, and quickly. Luckily you have been watching the recent developments in your game engine of choice and have learned that they have added a neat job system which can spread work over multiple cores and has a specialized compiler which can optimize the heck out of your code. “Yay” you say, but one limitation all your data need to be Blittable types or arrays of such so our previous ‘list of chunks of tiles’ approach isn’t going to work. Furthermore, it would be neat if the tile data was packed together in memory so you could write that jobs that traverse the whole set of tiles rather than jumping around memory to separate chunks. This becomes important as you watch your players and notice that, whilst they do drag out large chunks of tiles, they also place a lot of tiles one at a time.

Ok so let’s look at our old data layout

tile-data = [add(* * *), add(* *)]
undone-data = []

Now let’s make this a single array, and instead of undo removing one and adding to another we with have an index which we will call the active-head. It will point to the most recent active chunk of tiles

tile-data = [add(* * *), add(* *)]

Our active-head is represented by the v above tile-data in code it’ll just be an integer.

With that change undo looks like this

tile-data = [add(* * *), add(* *)]

and redo does this

tile-data = [add(* * *), add(* *)]

Cool, but now to remove those extra objects, let’s flatten the array and move the layout information to a separate array

tile-data = [* * * * *]
layout = [3 2]

So far so good. Now our layout tells us how many tiles are in each chunk and tile-data stores all the tiles packed together.

Undo now looks like this…

tile-data = [* * * * *]
layout = [3 2]

and redo is still the reverse. This means undo and redo are still as simple as changing an integer from the data-model side[3]

Delete is different though, delete is based on a region on the board so we need to check each tile to see if it intersects with that region and, if it does, then we remove it from the tile-data array. Undo naturally requires us to put that data back. Now from a data angle that is annoying as it could leave ‘holes’ in our array as in the following diagram (deleted tiles represented by x)

tile-data = [x * x x *]
layout = [3 2]

This is no good as now we have to handle this. We could have each tile have a flag that says if it’s alive or dead. We would then need to check that any time we scan over this array. This is ok but, unless we undo the delete, we still have memory being wasted as it’s occupied by data for tile that been deleted. So maybe we reuse the gaps? Sounds great, but we can’t fill in the gaps with tiles from newer operations as otherwise, we lose the property that tile-data is in ‘history order’ and thus lose the ability for undo/redo to be represented by the active-head.

So if we don’t want holes but can’t fill them then another option is to compact the array as we delete, so whenever tile data is deleted the tile data to the right is shifted left to fill in the gaps. This requires more copying during the delete but let’s assume we profiled it and it proved to be the best option[4]. This is what we will assume for the rest of the article.

Let’s now look at the above delete operation with compaction.

tile-data = [* *]
layout = [1 1]

Neat! but we should note we have complicated undo for ourselves. If we undo a delete We will have to ‘make room’ in tile-data for the old tiles to be copied back into. There is one blessing though, we only care that chunks are in history order, not the tiles within the chunks. That means we don’t need to make room in the same places as we had the tiles originally…

tile-data = [_ * _ _ *]
layout = [3 2]

Rather, we can make room at the end of each chunk…

tile-data = [* _ _ * _]
layout = [3 2]

And still keep all the undo/redo benefits we had previously with the active-head approach.

This also simplifies redo of a delete, can you spot how?

Originally we had to scan through the tiles to find the ones to delete. Redo requires us to delete those same tiles again but we know exactly where they are now, they are at the end of each chunk.

Chapter 2 - Multiple players and a quandary

Now things get ‘fun’. We can have up to 16 players on a board at a single time and all of them could be given the permissions to build. That means we need to keep that all in sync on all machines but all make sure everything feels instant.

Let’s get one thing out of the way quickly, order matters. By this I mean the obvious quality that an add followed by a delete can give a different result from a delete followed by an add, from this we know we need to apply changes in a fixed order. For us, that means when we build we will send a request to a ‘host’ which will decide what gets applied in what order and tells the other player’s games (called a client from now on). Note that we also need to see a change the moment it happens so the order will be:

  • Player tries to make a change
  • The local client applies the change immediately and sends a change request to the host
  • the local client listens for change acknowledgments from the host
  • if the acknowledgement is for the thing we just built then we are done
  • if the acknowledgement is for another client’s change then we undo our change, apply their change and then reapply our change

Now that is very briefly covered we will assume that things are applied in the same order on each client and ignore that aspect of the problem for the rest of the article.

To keep everything in sync we need to be able to send information about what is being done by each player. We already know that our selections could be very large so we don’t want to have to send information about every specific tile that was modified, so instead we will want to send just what happened e.g. ‘player 0 added 30 wall tiles in this area’. This is ok for us as we are already assuming a fixes order of operations so we can assume that every player has the board in the same state when they apply the change.

Multiple players building also gets interesting due to each player has their own undo/redo history. This is very important as shared histories get very annoying very quickly. However, this messes with our little data layout from before. In the following layout I have changed out the asterisks for a designation that tells us who added the tile, this is only for our benefit in this article, it needed to be encoded in the actual data like this.

tile-data = [p0 p0 p1 p1 p0 p0]
layout = [2 2 2]

So three chunks of tiles have been added. First player0 (p0) added 2 tiles, then player1 (p1) added 2 and then p1 added 2 more.

Now let’s say p1 hits undo, that means the middle two tiles need to be undone but we can’t do this just by shifting the active head.

Here we face two choices, either we drop the active-head approach or keep it and find a way around the above issue. Neither way is wrong and both require changes to how the data will be stored. For the sake of the article, I am just going to continue with the active head approach so we can see how this has to be modified to keep its nice properties in the face of multiple players.

Chapter 3 - Multiple Players with the active-head approach

OK, so let’s say we like the way we can just change a single integer to represent undo/redo using the active-head approach. What would be needed to keep this working with multiple players?

Well each player could have their own tile and layout data

p0-data = [p0 p0 p0 p0]
p0-layout = [2 2]
p1-data = [p0 p0]
p1-layout = [2]

This lets us regain the quality that undo for p1 only affects p1’s data.

From now on let us say that if a tile is in p0-data then player0 ‘owns’ that tile and if the tile is in p1-data then player1 ‘owns’ that tile. With that little bit of terminology, we can look at delete again.

Delete is based on a selection of tiles on the board, it does care about who ‘owns’ the tile behind the scenes; if the tile is in the selection, it will be deleted. That means that when we scan the data for tiles to delete we need to scan each player’s tile array.

Remember how undo of a delete needed to restore tiles to where they came from? Well, what do we do for tiles that we delete from other players? It turns out that adding them back to the other player’s chunks (and thus their history) results in behavior that is fairly confusing to play as the result of undo or redo depends to an extent on the current state of another player’s undo or redo. How robust a game feels depends a lot on to what degree a player’s expectations of game behavior match reality and so it’s sometimes better to have a simpler behavior even if a more correct but complicated option is available. I’m going to claim this is such a case without trying to prove it here :)

The result of this is that we don’t want to try to push undone deletes back into the data arrays of other players. So where do we put them? How about the end of our data section? This is a bad idea as is breaks the active-head approach again. So for us to continue using this approach we now need an additional store per player for undone deletes.

p0-data = [p0 p0 p0 p0]
p0-layout = [2 2]
p0-deletes = []

p1-data = [p0 p0]
p1-layout = [2]
= []

The addition of more per player data probably feels annoying (it did to me) but one upside is that this array can use the same active-head tricks that we have used for the player’s data array but with a slight difference, the tiles exist in-game only if they are to the right of the active-head. In this case:

  • a delete of tiles owned by other players pushes them onto the player’s deletes array and advances the active-head
  • an undo of the delete shifts the active-head to the left, the things to the right are now active in-game
  • a redo of the delete shifts the active-head right again


This might feel like it’s all getting a bit out of hand. That is a fair conclusion to make! This is why I said at the beginning this writeup is just an exercise in following an idea through to where it leads rather than making a call on the best approach. I hope you have some thoughts on this and also a feeling for where this kind of feature can get complicated.

I have left out a bunch of things like “how do we maintain the mapping between the tile data my game engine’s game objects when our data can only contain Blittable types?” and “How long can undo/redo histories be?”. They are fun in themselves as I do recommend considering them if you decide to try and come up with your own schemes.


[0] For example imagine player0 placing a block of tiles, then player0 deletes those tiles, then player1 places some tiles in that (now empty) spot. Finally player0 undoes their delete and suddenly player1’s tiles intersect with player0’s.

[1] This means we are ignoring both the fields within and also the various possible layouts for those fields.

[2] We could of course store the ‘bounds’ (a 3d region) of the tiles organized in a way that makes spatial queries easier, and then have those contain indexes back to the tiles they refer to. This will mean sorting/maintaining this structure as tiles are added/deleted, so it’s not a free lunch but you will want to look at this if you are building this yourself.

[3] you still need to handle the relationship between the data and actual GameObjects or entities (depending on the system you are using)

[4] remember that building only happens as fast as humans can click, whereas some things might be happening every frame (60 times a second). It can prove better to prioritize the more common case over the less frequent one even if it does result in a slight drop in performance for the less common case.

TaleSpire Dev Log 126

2019-09-03 10:31:48 +0000

Current dev work is focused on the new building systems.

My goal is to have the basics of the data model worked out locally by the end of the week. Most of this right now is focused on how best to handle undo/redo so that it’s fast but also simple to synchronize. I had previously designed had some schemes for this but some tools we’d like to use are currently still in preview and some would require significant engineering to work around (In this case driving rendering ourselves would require a lot of work on the lighting system) and we don’t have time for that.

@Ree has been getting back into prototyping the bit that players interact with, the tools and game feel of building. Progress is being made in both areas but nothing worth showing yet.

Thanks for dropping by! More updates soon.

TaleSpire Dev Log 125

2019-08-28 18:43:26 +0000

Whyyyyy, WHY did I say on discord that the next alpha update would be ‘soon’? It angered the demo-gods and we’ve been dealing with an asset-loading problem ever since.

That issue has now been fixed and I feel like we know when that update will be landing. However I am not evening hinting at when that could be given what wrath that might incur.

We have also been moving our assets out of Unity Collab. We have been using Collab for a while but there are aspects to it that make it painful for teams and there has been no sign that it will improve. For now we are using Git LFS which is what we have been using for the game itself for over a year. Git LFS, once set up, it’s been pretty good to us so far.

The money from the Kickstarter has landed and @Ree has been busy working on the business side of things. We are looking forward to both being able to say we are officially hired by Bouncyrock and also to give more little bits of info regarding what happens to money after you make your pledge.

I (Baggers) have just got back from a long weekend with family in the west of Norway. The last year and a half has been a lot of work so it’s really nice to be seeing folks again.

More news as it happens!

TaleSpire Dev Log 124

2019-08-15 23:40:30 +0000

It’s been a nice day today, both @Ree and I got some coding done. Here are a few highlights:

  • Dice feel regressed a while back and Ree got that fixed back up again.

  • He also updated the implementation of the radial menus which we hope has fixed the issues with some stats not showing. I tested this but we had never been able to get a reliable reproduction of the issue so we will see how this fix holds up in the wild.

  • I found a interesting default behavior in the networking library we used which was resulting in huge hangs when updating certain player information. Once found the fix was trivial.

  • We have enabled Unity’s incremental garbage collector to smooth over another GC spike we were seeing. It’s very much a band-aid but it’s a nice one for the alpha before we get to addressing the core issues behind the allocations.

More news and an update to the alpha in the near future.


TaleSpire Dev Log 123

2019-08-13 22:56:52 +0000

Day two of the dev planning and again things went well. The main job today was going back through the task lists and working out which ones would be dependent on which. This is important so we don’t get blocked by each other and that, for the cases where the is a dependency, there will be work the blocked person can be doing.

We wont be putting out these orders as they will change and we’d like to keep things flexible.

Following on from yesterdays work I also got the backend running fully without and internet connection so I’m hoping that with a little more work I can point Unity at my laptop and run the whole TaleSpire stack locally. This will be a big win for iteration time when developing the backend.

Lastly and most prettily @Dwarf has made an awesome looking Kobold which we will be seeing in game shots of in the near future.


That’s all for today,


TaleSpire Dev Log 122

2019-08-12 18:35:16 +0000

The end of last week saw me playing with a cool toy so I thought I’d write about it.

First though, I’ve carried on with my digging into various erlang bits and bobs.

With a bit of prodding I was able to get epmdless working with my erlang/docker setup. This is great as distributed erlang is one of the powerful features that many erlang libraries lean on and previously this was hampered by docker requiring explicit port mappings. If you are interested in epmdless in conjunction with docker you can check out this article here.

I also spent some quality time getting to know gproc, erlbus, websockets a little better and I’m much happier now with where I’d use them in my projects. I’ve not got much more concrete to say on these but having a good understanding of common tools is helping with the design of the next iteration of the backend (more news on this in future too).

Lastly as fun bit (for me). For a time I was really enjoying that I was able to use docker to test the TaleSpire backend locally. I had webservers, db servers etc on their own little network and I could get a reasonable idea of how things should run. However at one point we started using amazon’s S3 for storage and that put a spanner in the works as I couldn’t test that locally. I could make a alternative approach for local dev but that’s more code to maintain and that could fail in ways that confuse my testing. Luckily there is a project called minio which runs in a docker container and presents the same api as S3. I’ve already modified the image to include my preferred tweaks and have tested making presigned urls which work GREAT.

So now I get to go remove some ‘local dev’ hacks and I get a simpler, more realistic, fully local, dev environment. LOVE IT.

Alright, that’s all the fun nerdage for now, back to planning :)

p.s. One thing I got stuck on initially with epmdless was that I was using rebar3 shell in the entrypoint script for my container. This will not work with epmdless as the epmd module vm args would need to be passed to shell and that would then freak out as it wouldnt know where to find the epmdless module. So instead try having the entrypoint script make a debug release of your erlang app and then start that in foreground mode. Then your config/vm.args will be used and everything will work fine.

Also, remember to expose the remsh port for epmdless so you can use the remote shell. I used this for my erlang repl settings in emacs. Once you have the official examples running most of this should be fairly self explanatory.

(setq inferior-erlang-machine-options
      '("-env" "EPMDLESS_DIST_PORT" "18999"
        "-env" "EPMDLESS_REMSH_PORT" "18000"
        "-name" ""
        "-setcookie" "cookie"
        "-remsh" ""
        "-proto_dist" "epmdless_proto"
        "-epmd_module" "epmdless_client"
        "-pa" "/app/_build/default/lib/epmdless/ebin/"))